Waveform Classification by Memristive Reservoir Computing

Gouhei Tanaka, Ryosho Nakane, Toshiyuki Yamane, Seiji Takeda, Daiju Nakano, Shigeru Nakagawa, Akira Hirose

Abstract: Reservoir computing is one of the computational frameworks based on recurrent neural networks for learning sequential data. We study the memristive reservoir computing where a network of memristors, instead of recurrent neural networks, provides a nonlinear mapping from input sequential signals to high-dimensional spatiotemporal dynamics. First we formulate the circuit equations of the memristive networks and describe the simulation methods. Then we use the memristive reservoir computing for solving a waveform classification problem. We demonstrate how the classification ability depends on the number of reservoir outputs and the variability of the memristive elements. Our methods are useful for finding a better architecture of the memristive reservoir under the inevitable element variability when implemented with nano/micro-scale devices.